6781097121

EPIA'2011

ISBN: 978-989-95618-4-7

If the player folds 72% or inore hands then he is considered to be a tight player else he is loose player. Regarding the aggression factor (AF), if the player has an AF above 1 then he is aggressive, else he is considered to be passive. Through this simple classification it is possible to know which hands will be likely played by each opponent.

In these articles [1,2], the concept of Hand Strengtli and Hand Potential, tliat was previously defined here [7], was slightly modified to support opponent modeling. Hand strength is the probability that a given hand is better than any other possible hand. Instead of generating all possible two card remaining hands, the authors suggest that one should only generate the hands with cards that the opponents likely have. The same idea could be applied to the Hand Potential algorithm, which is the probability of a given hand improves to be the best at the finał round.

Despite all the breakthroughs achieved by known research groups like Computer Poker Research Group¹ and individuals, no artificial poker playing agent is presently known to be capable of beating the best human players.

To the datę there are very few approaches that use clustering algorithms to define types of coinmon strategies. This article attempts to fili this gap, by defining new types of strategies which are represented by the freąuency of certain actions. The clustering algorithms are used on a very large database in order to extract common playing styles from its players.

3.1 Clustering algorithms

Clustering algorithms are imsupen ised leaming methods that divide a set of objects into several subsets (called clusters) where the objects on the subset are in some way related. Clustering has multiple applications in areas such as machinę leaming. data mining, imagine analysis and others [8], A simple 2D axis clustering can be seen on Figurę 2.

Fig. 2. Clustering example

In this figurę, a chart is presented with several points. There are clearly two groups of points that can be identified by the proximity between themselves (blue and red groups). This example is very simple, only presenting two dimensions. Clustering can be applied to problems with multiple dimensions that aren t so easily graphically represented.

One particular type of clustering algorithms and the one that is used on this work is partitioning algorithms. This type of algorithms creates various partitions of objects at

Computer Poker Research Group from University Alberta: http://poker.cs.ualberta.ca/